"All the market return comes from just 4% of shares"

[SPC100]

Irishtimes have an article about the research today

Most stocks are flops: just 1% of stocks account for all market gains

Picking the handful of stocks that deliver the bulk of investor returns is seeking a needle in a haystack

www.irishtimes.com

This looks to be the relevant original research papers for those interested in reading more

Do Stocks Outperform Treasury Bills?

Four out of every seven common stocks that have appeared in the CRSP database since 1926 have lifetime buy-and-hold returns less than one-month Treasuries. Whe

papers.ssrn.com

Do Global Stocks Outperform US Treasury Bills?

We study compound returns to nearly 62,000 global common stocks during the 1990 to 2018 period, documenting that the majority, 56% of US stocks and 61% of non-U

papers.ssrn.com

 

[Colm Fagan]

@SPC100 Thanks for providing the links to the two articles mentioned. I've downloaded them and hope to study them shortly.

I confess to an element of scepticism (based partly on my own experience of running a highly concentrated portfolio for almost two decades now). I will study them with a jaundiced eye.

A couple of other reasons for being sceptical:
1. Earlier in this thread, @Sarenco alerted us to what is now called the "Voya Corporate Leaders Trust Fund". This fund has been in existence since 1935 and has stuck with exactly the same stocks, unless they went bankrupt. Only 22 of the original 30 are still standing. All the money is in those stocks (allowing for the fact that some have been taken over, etc.). The performance of that fund exceeded the return on the S&P 500 over the 40 years to 2016. This is despite the fact that it would have missed out completely on all the glamour stocks - Microsoft, Amazon, Apple, etc. It didn't even include IBM! Does that square with this academic's conclusions?
2. I looked at another academic study some time ago, which arrived at similar results to those quoted in the Irish Times article. As I recall, it gave exactly the same weighting to a fly as to an 800-pound gorilla, which is nonsense. Most small companies go bust in their early years. I'm sure much the same is true for listed stocks. To be true to real life, the study should allow for relative size and should also limit the universe to stocks that have been quoted for (say) five years.

A final comment is that such papers help to convince ordinary investors not to trust their own judgement when buying stocks. I'm not surprised that an active asset manager gave financial support to one of the authors, but maybe that's just my suspicious mind!

Anyway, thanks again for posting the references to the papers, and I look forward to reading them.

 

[Sarenco]

Here's a link to the Vangurad paper referenced in the IT article (which is worth a read IMO) -

 

[Sarenco]

@SPC100 The performance of that fund exceeded the return on the S&P 500 over the 40 years to 2016. This is despite the fact that it would have missed out completely on all the glamour stocks - Microsoft, Amazon, Apple, etc. It didn't even include IBM! Does that square with this academic's conclusions?

Interesting question.

I think the original stock selection of the "sloth fund" is important. The fund's promoters invested equally in the common stocks of the 30 biggest US companies (excluding financials) back in 1935. Their central thesis was that if a company could survive the Great Depression, it could survive anything. History has largely proved them right.

To put it another way, an extreme "quality screen" was applied to each stock selected. If a company suspended dividend payments, it was dumped.

These were all companies with large competitive "moats". The stock selection followed a formula - it was by no means random or subjective.

It is certainly true that the sloth fund has (modestly) outperformed the S&P500 over the last 40 years. However, that is largely due to the fact that the fund holds practically no tech stocks and therefore avoided the worst of the 2000 tech crash. Conversely, the sloth fund has materially lagged the S&P over the last 10 years (largely because it doesn't hold any FAANGS).

Will this outperformance be repeated over the next 40 years? That's obviously possible but I have my doubts.

So, it's clear that the sloth fund is something of a poster child for the long term benefits of a buy and hold approach, while limiting transaction costs.

Does it jar with the academic research that indicates that market returns are generated by a small minority of stocks? Well, I don't think so.

Incidentally, I entirely agree with your 2nd point regarding the market capitalisation of the universe of stocks under consideration. I tried to make that exact point (albeit far less eloquently) earlier in this thread.

 

[Brendan Burgess]

A final comment is that such papers help to convince ordinary investors not to trust their own judgement when buying stocks. I'm not surprised that an active asset manager gave financial support to one of the authors, but maybe that's just my suspicious mind!

I haven't read the papers yet, but I don't understand this point Colm.

If all the returns come from 1% or 2% of the shares, would it not make it even more difficult for active fund managers to outperform?

If I believed these figures, then surely I should buy an ETF? I need a few hundred shares to make sure I get the very few successful ones.

Brendan

 

[Colm Fagan]

it's clear that the sloth fund is something of a poster child for the long term benefits of a buy and hold approach, while limiting transaction costs

Hi Sarenco. Firstly, thanks for an excellent post. I've picked out this quote, because it supports my basic contention that someone saving for retirement over (say) 20 years and then drawing down those savings over (say) the following 20 years is best advised to stick the money in a small number of good quality stocks and leave it there, reinvesting the dividends in the run-up to retirement, then taking the dividends plus gradual share sales for “income” in retirement.
This is a vastly superior strategy to the conventional wisdom from financial advisers, wealth managers and insurance companies that they should invest in a mix of equities/property, bonds and cash in the run-up to retirement, shifting gradually to less “risky” investments as retirement date approaches, and then either buy an annuity (i.e. put everything in bonds) or reduce the equity proportion substantially in the final 20 years, i.e. throughout their retirement. The question of whether the long-term performance of their chosen stocks is slightly better or slightly worse than "the market" is of little significance relative to the high-level equity/bond/cash asset allocation question.

And this conclusion is even before allowing for asset management fees and other charges. From what I've seen, some of the charges on post-retirement products are quite excessive.

If all the returns come from 1% or 2% of the shares, would it not make it even more difficult for active fund managers to outperform?

If I believed these figures, then surely I should buy an ETF? I need a few hundred shares to make sure I get the very few successful ones.

Brendan, I see your point, but I think the authors (more likely, their asset manager sponsors) are trying to frighten off DIY investors, to convince them that they need “professional” help in choosing investments. My views on this subject are set out in the diary entry “A guy in the attic”.

 

[Duke of Marmalade]

Here's a link to the Vangurad paper referenced in the IT article (which is worth a read IMO) -

I read it. Mostly common sense and not really needing 10,000 simulations.
The glaring mathematical fallacy of Figure 5 rather devalued the whole paper for me. And it is fairly central to the message. For example it says the average expected excess return (AEER) for a 1 stock random portfolio is -9.9% compared to a fully diversified approach. But the math of the simulation algorithm would dictate that the AEER should be zero no matter what size of portfolio.
So where is the error? The paper splits the simulations into those with positive excess return which for the 1 stock portfolio occurs on 11.1% of occasions with a conditional average excess return of +4.2%. Conditional average excess returns on the 88.9% simulations which underperfomed was -11.7%. The paper then gets the weighted average of these figures which is indeed -9.9%. But that is false math. What should have been done is to find the weighted average of the two returns rolled up for 41 years and then find what the equivalent annual return is over those 41 years. Within rounding the AEER at all portfolio sizes is, as the math would dictate, zero.

Another piece that stuck in the craw:

We found that less diversified portfolios have more relative risk than more diversified ones and that investors should therefore expect higher returns from the less diversified portfolios, but the evidence from our simulations shows that concentrated portfolios have lower average returns than diversified portfolios.

Besides the fact that (see above) the evidence does not show this last at all, they really do show a basic misunderstanding of financial theory here. Risk is not rewarded as a moral imperative!! Of course concentrated portfolios are more risky than diversified ones but, so the theory goes, this risk can be avoided (by diversifying). There is no basis whatsoever for expecting a risk that can be avoided to be rewarded by the market.

 

[Colm Fagan]

I don't know if readers of this forum realise how lucky we are. Vanguard is one of the most highly regarded asset management companies in the world. It works hard to maintain its reputation. I presume this paper went through a thorough internal peer review process and that its conclusions are being quoted as gospel worldwide, yet here is one of our own demolishing a couple of its key arguments. Thanks @Duke of Marmalade !

 

[Sarenco]

Of course concentrated portfolios are more risky than diversified ones but, so the theory goes, this risk can be avoided (by diversifying). There is no basis whatsoever for expecting a risk that can be avoided to be rewarded by the market.

Very fair point Duke.

Still, I thought the simulated portfolios' probability of outperforming the (equal weighted) benchmark, based on historical data, was interesting nevertheless. For example, the finding that a (randomly selected) 30 stock portfolio had a 40.3% probability of outperforming the benchmark is much closer to a coin toss then I would have predicted.

 

[Duke of Marmalade]

Very fair point Duke.

Still, I thought the simulated portfolios' probability of outperforming the (equal weighted) benchmark, based on historical data, was interesting nevertheless. For example, the finding that a (randomly selected) 30 stock portfolio had a 40.3% probability of outperforming the benchmark is much closer to a coin toss then I would have predicted.

Sarenco, this paper and in particular Figure 5 really got me thinking (so yes it was worth a read), in fact it got me doodling on Excel. The latter part of the paper showing, for example, that a 52% success rate (as defined) in picking stocks gives fairly significant out performance expectations surprised me and may be of some interest, though I haven't subjected the methodology to much scrutiny.
But the whole earlier part of the paper is vacuous. All the simulations are showing is what are statistical truisms. The results are in no way dependent on sampling historic data. Figure 2 and the fact that outliers dominate the benchmark is not necessary to come up with these truisms. If we sampled from a perfectly well behaved data set where all stocks show the same distribution we would still find that the smaller the portfolio size the greater the dispersion. More importantly, whilst the expected overall return is the same for all portfolio sizes the probability of being higher than the expectation is lesser for the smaller portfolios. But since it is an expectation there is clearly a compensation for having less chance in achieving your expectation in having a higher chance of shooting the lights out.
I repeat that these observations would be true no matter what distribution you chose for the population you are sampling from, they are totally independent of past data.
Just to get a bit wonkish here, the reason for these truisms is the Central Limit Theorem. In effect the methodology is producing a Lognormal distribution of the final returns and from this the truisms follow. The Central Limit Theorem essentially smooths away any idiosyncracies in the underlying data set that is being sampled.

 

[Sarenco]

If we sampled from a perfectly well behaved data set where all stocks show the same distribution we would still find that the smaller the portfolio size the greater the dispersion.

Sure but all stocks didn't show the same distribution so surely the historic data impacts the degree to which the dispersion narrows as the number of stocks increases. No?

 

[Duke of Marmalade]

Sure but all stocks didn't show the same distribution so surely the historic data impacts the degree to which the dispersion narrows as the number of stocks increases. No?

Yes, that is true. But if the Central Limit Theorem applied in full the only parameter of the dispersion that matters would be the variance. And yes outliers do increase variance, but the skewness they impart is planed away by the CLT.
In this particular example the CLT is not fully applicable as the different quarterly rebalancings are not from iids. Nonetheless I would suggest that the variance dominates the dispersion of the final outcomes.
I wonder is it possible to access the Vanguard database.

 

[RobFer]

I don't know if readers of this forum realise how lucky we are. Vanguard is one of the most highly regarded asset management companies in the world. It works hard to maintain its reputation. I presume this paper went through a thorough internal peer review process and that its conclusions are being quoted as gospel worldwide, yet here is one of our own demolishing a couple of its key arguments. Thanks @Duke of Marmalade !

I understand there is good reason to be sceptical but there is a hierarchy in finance research too.